Make help message template configurable

[dlqqq]

trafficstars

Description

• Makes the help message template configurable via traitlets.
• Performs some fairly significant refactoring. The method for sending a help message has been moved from HelpChatHandler to the BaseChatHandler base class.
  • The reason for doing this is that /clear also needs to be able to send a help message. Since this functionality is shared by multiple chat handlers, it seemed like better practice to implement this exactly once in a base class and allow any chat handler to send a help message.
• Fixes #932

Demo

Testing

To reproduce the above demo:

1. Create a new config.py file in your current directory with the contents:

c.AiExtension.help_message_template = """
Sup. I'm {persona_name}. This is a sassy custom help message.

Here's the slash commands you can use. Use 'em or don't... I don't care.

{slash_commands_list}
""".strip()

2. Start JupyterLab via jupyter lab --config=config.py.

Additional notes

I've added a new TestProviderAskLearnUnsupported class in the jupyter_ai_test package for local testing. You can verify that after switching to this model in the settings and running /clear, /ask and /learn do not show up in the help message as they are listed in unsupported_slash_commands.

• We do not automatically regenerate the help message when switching between different LLMs, even if they differ in the slash commands that they support. This is a known issue.

[michaelchia]

hello, since you are making this change, it seems like perhaps we can resolve https://github.com/jupyterlab/jupyter-ai/issues/851 by splitting the welcome and help message. I think it would make more sense to make the welcome message configurable via this traitlet while having the help chat handler define the help message. what do you think?

[dlqqq]

@michaelchia I agree that we should split the welcome message and the help message in the future! This PR doesn't exclude that possibility; it's merely making the help message configurable.

[krassowski]

I think this also fixes https://github.com/jupyterlab/jupyter-ai/issues/927